[171]*171EDMONDS, P. J.
Defendant has petitioned for reconsideration of our opinion affirming his convictions for robbery and kidnapping. State v. Lytsell, 184 Or App 75, 55 P3d 503 (2002). We allow the petition, withdraw our former opinion, and reverse those convictions.1
We restate the facts as we described them in our previous opinion:
“Defendant and two other men, Sandborn and Clippinger, burglarized the home of the victim, Fornataro, on the evening of January 5, 1998. While the three intruders were inside the house, Fornataro and his son returned home. Fornataro testified that he became suspicious when he noticed that lights were on and that items in the house had been disturbed. He went to check his bedroom and, as he entered, he saw two men. One man was rummaging through a cabinet in the room, and his upper torso was hidden from view by the cabinet’s doors. The other man was standing about seven feet away, holding Fornataro’s handgun in his left hand. As Fornataro entered the room, he asked the two men what they were doing. Startled, the man holding the gun pulled out another gun and pointed it at Fornataro. The gunman then told Fornataro to turn around or he would shoot him. Pressing the gun against Fornataro’s neck, the gunman forced him outside onto the deck. Fornataro testified that, once outside, the gunman ordered him to get down on his knées, but Fornataro refused. The gunman and the second intruder then fled through the backyard and over a fence. Fornataro also saw the profile of a third intruder who left the house through a basement window. After the intruders left, Fornataro called the police.
‘While investigating the crime scene, the police found a pager that belonged to defendant in Fornataro’s backyard. In addition, a police artist drew a composite facial sketch of the gunman from Fornataro’s description. The sketch strongly resembled defendant. The police also conducted three photo throw-downs. In the first throw-down, Fornataro identified another man, rather than defendant, [172]*172as the gunman. That throw-down did not include defendant’s photograph. The second throw-down included defendant’s photograph, and Fornataro identified him as the gunman. In a third throw-down, which also did not include defendant’s photograph, Fornataro commented that the person in one of the photos resembled the gunman, but he told the officer that he was certain that it was not the gunman.
“The police eventually arrested defendant for the charged offenses. Sandborn also was in custody but later escaped; at the time of defendant’s trial, Sandborn had not been apprehended.
“Defendant waived his right to a jury trial, and the case was tried to the court. At trial, defendant admitted that he was one of the burglars but denied being the gunman. He testified that he, Sandborn, and Clippinger were in the bedroom when Fornataro entered. According to defendant, when Fornataro entered, defendant was crouched behind the bed looking underneath it, Clippinger was searching through a cabinet, and Sandborn was standing beside the bed. As soon as Fornataro confronted them, Sandborn pulled the gun out and ordered Fornataro to turn around. Sandborn then ordered Fornataro to lead them outside onto the deck. However, when Fornataro turned around, Clippinger jumped out the bedroom window. Defendant followed Sandborn and Fornataro through the house, but he eventually fled through another bedroom window. Thus, according to defendant, Sandborn was the only person on the deck with Fornataro. After fleeing, defendant and Sandborn went to a nearby house that belonged to Cannon, a friend of defendant’s mother. Defendant’s teenage stepsister, Boyd, was at the house with several of her friends, including Byles. When defendant and Sandborn arrived at Cannon’s house, Boyd let them in and the two men ran down into the basement. Defendant stated that, after cleaning up, he and Sandborn returned upstairs and talked with the girls.
“Fornataro testified that, during his encounter with the gunman, he had carefully studied the gunman’s face for the specific purpose of later identifying him. Then, in open court, he identified defendant as the gunman. Fornataro also described his thoughts during the photo throw-downs. [173]*173He explained that, in the first throw-down, he had identified a man whose facial features were similar to defendant’s, but he told the officer that the person in the picture was thicker in the neck and torso than the gunman. He stated that he had quickly identified defendant as the gunman during the second throw-down, and, in the final throw-down, he noticed a man who looked similar to the gunman but, even at that time, he had been ‘absolutely sure [that] it was not the gunman.’
“Defendant called as a witness the police officer who conducted the first two throw-downs. The officer testified that Fornataro had identified a suspect in 38 seconds during the first throw-down and had identified defendant in 2 minutes and 35 seconds during the second throw-down. Furthermore, according to the officer, Fornataro mentioned during the second throw-down that another man somewhat resembled the gunman.
“Near the end of defendant’s case-in-chief, and after defendant testified, he sought to introduce in evidence hearsay declarations allegedly made by Sandbom that implicated Sandborn as the gunman. Defendant made an offer of proof that included the in-court testimony of three witnesses. Two of the witnesses, Boyd and Byles, testified concerning statements Sandbom allegedly made when he and defendant were at Cannon’s house on the night of January 5. Boyd testified that Sandbom told Boyd and Byles that he had held the gun to Fomataro’s head so that he, defendant, and Clippinger could escape. Byles testified that she heard Sandborn tell the group that Tie had made a guy get down on the ground and held [the gun] up to the guy’s head.’ However, she later testified that she did. not remember Sandbom’s exact words, only that “he said — told the guy to get down — :down to the ground.’ The third witness, Lohr, testified that several months before defendant’s trial, but after defendant had been arrested, Sandborn told her that “he feels bad because [defendant is] going to be doing his time for him.’
“Defendant argued that the declarations were admissible as statements against Sandborn’s penal interest under OEC 804C3XC)1 and that the declarations exonerated defendant of any crimes involving the use of the gun. Defendant contended that the statements were sufficiently corroborated because the two witnesses heard ‘relatively the same [174]*174statement regarding the gun under the same, circumstances.’ Furthermore, defendant contended that the statements were consistent with the events that occurred during the episode and that Sandborn voluntarily made the statements in surroundings in which he felt comfortable. Defendant urged that Sandborn’s comment about defendant doing ‘his time’ was against Sandborn’s penal interest, and that it also corroborated Sandborn’s statements to Byles and Boyd.
“The state conceded that the declarations allegedly made to Byles and Boyd constituted statements against penal interest, but it argued that the declaration that defendant was going to be ‘doing his time’ was not against Sandborn’s penal interest. The prosecutor also argued that all of the statements were inadmissible because they lacked sufficient corroboration or indicia of reliability. Specifically, the state focused on an inconsistency in the evidence: Although Byles testified that Sandborn told her and Boyd that he had forced Fornataro to the floor, Fornataro testified that he had refused to comply with the gunman’s demand. In addition, the state pointed to the close relationship between the witnesses and defendant and suggested that bias in favor of defendant may have motivated them to lie.
“At the conclusion of defendant’s offer of proof, the trial court excluded the proffered evidence. It reasoned:
“ ‘The only question before me at this time is the admissibility of the statement. [The state] pointed out, and I previously found, that [Sandborn] is unavailable as defined by the evidence code.
“ ‘There are two other, criteria that I think have to be met. First, the statement has to be a statement against penal interest. And I do find that the statement of [Sandborn] is a statement against penal interest. That alone is not enough to have it come into these proceedings.
“ ‘The Court in [State v. Thoma, 313 Or 268, 834 P2d 1020 (1992),] said there needs to be some corroborating evidence. Or to say it a different way, beyond a statement against penal interest there must be some assessment as to the reliability of the proffered evidence.
[175]*175“ ‘The Thoma court specifically said that they were not deciding on what unavailable — excuse me, what criterion would necessarily corroborate any statement, but there are seven things that they made reference to, and I would like to contrast or compare to this case.
“ ‘There is a connection between the defendant and the declarant.
“ ‘There is a known connection between the declar-ant, the defendant and all of the witnesses.
“ ‘The statement was made on a single occasion. And in making that finding, I am finding that the statement that the defendant would be doing his time is so vague as to not be corroboration. And I’m finding it vague for this reason. If the defendant is, in fact, convicted of burglary and no other charge, and if the evidence I have before me is accurate, and if [Sandborn] is never caught, it could still be said that the defendant is doing [Sandborn’s] time.
“ T further find that the statement of [Sandborn] is factually inconsistent with the testimony that’s presented in this case in one particular important element, and that is the statement about whether or not he made the victim get down on the ground. The testimony in this case by the victim, and it has not been contradicted by the testimony of defendant, is that a gun was pointed to his head, he was walked outside, he was told to get down on the ground, he refused to get down on the ground and, in fact, — and I don’t have my — that page of my notes in front of me, but said something to the effect that if you’re going to shoot me, you’ll have to shoot me standing up.
“ ‘So as I analyze the corroboration and the corroboration that’s been argued by [defendant], I come to one conclusion, and that’s the statement against penal interest and that’s the only conclusion I can come to. I do not find that there’s sufficient corroboration and the statements of [Sandborn] will be excluded and will not be considered by me.’
“After the court excluded the statements, defendant called three additional witnesses. The first witness, Stickley, testified that she previously had heard Sandborn use a distinctive expression involving the use of a gun. The [176]*176second and third witnesses, Lohr and Boyd, identified a photograph of Sandborn. In rebuttal, the state recalled Fornataro, who reiterated that defendant was the gunman and testified that the man in the photograph identified by Lohr and Boyd was not the gunman.
“At the conclusion of the trial, the court found defendant guilty of each of the charged offenses. The court explained its ruling as follows:
“ “When [the victim] first testified, I did with him what I did with every witness, I made an initial evaluation whether I thought he was credible or not credible. Having made that initial determination, I weighed that testimony against every other witness’s testimony, and, I think, reevaluated whether I thought he was credible or not credible. That’s not true just of [the victim], it’s true of every witness who testified.
“ ‘I’m troubled by one thing. There are a number of things that cause me to reach my decision. I found [the victim] initially to be a credible witness. And in finding him to be a credible witness, I found his initial account to be credible. When the defendant testified, the defendant put himself in a position that his head was under the bed, that [the victim] could not, should not have been able to see him, and there’s nothing the defendant said that changed that.
“ ‘[Defense counsel], your argument that his picking [defendant’s] picture out of a photo throw-down because the throw-down is overly suggestive, I do not find that it is. That he was the only one who had a threatening look on his face or that it was in the number one position does not give me a satisfactory explanation of why he made the identification. I do find that [the victim’s] statement that he studied the face of the gunman for the specific purpose of making an identification later to explain why he made that identification.
“ ‘There are a number of other things that I considered in reaching my decision, but I do find the defendant guilty of all of the charges against him.’
“On appeal, defendant assigns error to the exclusion of Sandbom’s statements. He argues that the statements were admissible under OEC 804(3)(c). Moreover, defendant asserts that the error was not harmless because the trial [177]*177court did not consider the statements in reaching its verdict, and they were critical to the defense theory that Fornataro had misidentified him as the gunman. The state responds that Sandborn’s statements were not admissible and, regardless, any error was harmless.
"1 OEC 804 provides, in part:
“ ‘(3) The following are not excluded by ORS 40.455 if the declarant is unavailable as a witness:
a
“ ‘(c) A statement which * * * at the time of its making * * * so far tended to subject the declarant to * * * criminal liability * * * that a reasonable person in the declarant’s position would not have made the statement unless the person believed it to be true. A statement tending to expose the declarant to criminal liability and offered to exculpate the accused is not admissible unless corroborating circumstances clearly indicate the trustworthiness of the statement.’ ”
Lytsell, 184 Or App at 77-83.
In our former opinion, we did not decide whether Sandborn’s statements were admissible. Instead, we held that, even if they were, the trial court’s decision to exclude them was harmless error. We now conclude, for reasons that we describe later, that any error was not harmless. We must therefore first determine whether the trial court erred in refusing to admit the evidence; if it did not, there is no issue of harmlessness.
We agree with the trial court and the state that Sandborn’s statements to Byles and Boyd were against his penal interest, while his statement to Lohr was not. In order for those statements to be admissible under OEC 804(3)(c), corroborating circumstances must clearly indicate their trustworthiness. The appellate courts have not given a thorough discussion of the corroboration requirement under OEC 804(3)(c), see Wood v. Baldwin, 158 Or App 98, 104, 972 P2d 1221, rev den, 329 Or 61 (1999), and we need not do so here. On appeal, the parties emphasize two issues: first, whether [178]*178the trial court could consider the credibility of the witnesses — Byles and Boyd — who testified that Sandborn made the statements, and second, whether the other circumstances are sufficient to serve as corroboration.
We agree with the state that the trial court could consider the credibility of Byles and Boyd. Although the language of OEC 804(3)(c) does not answer the question of whether the trial court may consider their credibility, the legislative commentary to the rule makes the legislature’s intent clear. The language of the rule is ambiguous because its requirement that corroborating circumstances indicate “the trustworthiness of the statement” could mean either that only the statement of the hearsay declarant requires corroboration or that the evidence that the declarant actually made the statement also requires corroboration. The legislative commentary resolves the ambiguity. It states:
“The common law refused to concede the adequacy of [statements contrary to] penal interest in large part because it distrusted evidence of confessions by third persons offered to exculpate the accused. This reflected a suspicion that either the fact or the contents of the confession were fabricated, enhanced in either case by the required unavailability of the declarant: To meet that concern, [OEC] 804 requires that a statement against penal interest be corroborated.”
Reprinted in Laird C. Kirkpatrick, Oregon Evidence § 804.04[2] (4th ed 2002) (emphasis added). The legislature’s concern, thus, was to avoid fabrication of the fact of the statement as well as of its contents. Thus, the trial court could appropriately consider evidence that relates to whether Sandborn actually made the statements as well as to whether it is otherwise trustworthy.
In this case, the only part of the trial court’s analysis that might implicate Byles’s or Boyd’s credibility was its comment that there was a connection between them, Sandborn, and defendant. However, its discussion of that point focuses on the trustworthiness of Sandborn’s statements, not on whether Byles and Boyd accurately reported what he said. In Thoma, which the trial court cited in its discussion, the Supreme Court relied in dictum on the absence of a known [179]*179connection between the defendant and the declarant to support its conclusion that there was sufficient corroboration for the declarant’s statements.2 313 Or at 278-79. There was no dispute in Thoma about whether the declarant had made the statements in question, so the lack of a connection related to the trustworthiness of the statements, not to whether they were made. The trial court in this case did not suggest, in explaining its ruling, that the connection between Sandborn, defendant, and Boyd and Byles undercut their credibility as witnesses.3 In the rest of its discussion, it assumed that Sandborn had in fact made the statements. We accept those conclusions and turn to the issue of the sufficiency of the corroborating circumstances.4
A number of facts tend to corroborate Sandborn’s statements. First, he made them almost immediately after the events to a group of people whom he had reason to trust. His action could well have been in part an explanation of why defendant and he had made so hurried an appearance and in part a way of releasing the tension from having been surprised during the burglary and having made so dramatic an escape. Second, Sandburn did not make the specific statements at issue in isolation. Rather, they are part of Sandborn’s general description of what had happened that evening and of his role in it. The specific statements come naturally from the flow of his discussion. At the time, Sandborn did not know that the burglary would be discovered, that he would make himself unavailable, and that defendant would be tried for things that, according to the statements, Sandborn had done. Thus, at the time there was no apparent reason for him to think that what he said would [180]*180benefit defendant or anyone else. Third, Fornataro quoted the gunman as using a phrase that Sandborn, but not defendant, frequently used. According to Fornataro, the gunman threatened him, saying, “I’ll cap your ass, nigger.” Witness Stickley testified that, while she had often heard Sandborn use that phrase, she had never heard defendant use it. Fourth, the statements are generally consistent with Fornataro’s description of what happened, aside from Byles’s uncertainty about whether Sandborn said that he made the “guy” get down or only told him to do so.5 Finally and most importantly, the police found physical evidence that corroborates defendant’s position that Sandborn, rather than he, was the gunman. Defendant’s pager was seized from underneath the window of a different bedroom from the one where Fornataro surprised the burglars. It was not found near the deck where the gunman took Fornataro. That fact supports defendant’s testimony that he fled through that window, which would mean that he was not the gunman.6
There is no magic formula to the corroboration requirement. Each case will have its own particular circumstances. The legislature has told the courts to construe the requirement “in a manner that effectuates its purpose to circumvent fabrication.” Legislative Commentary to OEC 804(3)(c), reprinted in Kirkpatrick, Oregon Evidence § 804.04[2], There are significant indicia of reliability in Sandborn’s statements as explained above, and the surrounding circumstances are also consistent with its accuracy. The corroboration is sufficient for the statements to be admissible, and the trial court erred in excluding them.7
[181]*181We turn to whether the trial court’s error affected defendant’s substantial rights — that is, whether or not it was harmless.8 OEC 103(1)(b).9 Defendant, relying on State v. Joslin, 332 Or 373, 387, 29 P3d 1112 (2001), states the standard for harmless error as whether there is “ ‘(1) substantial and compelling evidence of guilt * * *, and (2) little, if any, likelihood that the error affected the verdict.’ ” (Quoting State v. Walton, 311 Or 233, 230, 809 P2d 81 (1991).) He argues that that standard does not permit the reviewing court to assess the credibility of witnesses or to evaluate the evidence. The state, relying on State v. Cunningham, 179 Or App 359, 382, 40 P3d 1065, adh’d to on recons, 184 Or App 292, 57 P3d 149 (2002), rev pending, (2003), states the test as involving two questions: “First, what was the relative strength of the parties’ evidence? And second, in the totality of the parties’ evidence, how significant was the erroneously [excluded] evidence?” It argues that that test involves evaluating the quality of the evidence admitted and excluded.
Contrary to defendant’s apparent assumption, Joslin and Cunningham do not state separate tests; rather, Cunningham breaks the second portion of the test in Joslin— the likelihood that the error affected the verdict — into two parts. As the substantive discussion in Cunningham indicates, applying that test necessarily involves evaluating the relative strength of the state’s case as compared to the defendant’s case both with and without the challenged evidence.10 The purpose of harmless error analysis is to determine whether the evidence in question could have affected the ultimate verdict; if not, the error was harmless. It is difficult to conceive how we could make that determination without attempting to determine how the trier of fact was likely to evaluate the evidence as a whole.
[182]*182In this case, there is substantial evidence from which a trier of fact could find that defendant was the gunman: he was admittedly one of the burglars, and Fornataro clearly and unequivocally identified him as the gunman. The trial court relied primarily on that identification in convicting defendant, noting that, at the time of the burglary, Fornataro carefully observed the gunman in order to be able to identify him later. In our previous opinion, we held that the evidence of Sandborn’s statements would not have affected that conclusion. We emphasized that the composite sketch that a police artist drew, based on Fornataro’s description, shortly after the burglaiy strongly resembled defendant and that Fornataro’s positive identification of defendant in the second throw-down and at trial contradicted defendant’s testimony that he was hidden behind the bed when Fornataro entered the bedroom. We stated that, in contrast, defendant’s version of events was improbable. Lytsell, 184 Or App at 85. The evidence came from defendant’s close friends and a relative and, more significantly, it did not explain how, in light of defendant’s testimony of the physical facts, Fornataro was able to identify defendant so credibly or why he insisted that Sandborn was not the gunman. Thus, we concluded, the excluded statements would not have been significant to the trier of fact. Id. at 86.
On reconsideration, we conclude that we did not adequately evaluate the evidence as a whole. In particular, we did not recognize the questions that the record as a whole presents concerning Fornataro’s identification of defendant as the gunman and the way in which Sandborn’s identification of himself as that person could have changed the evaluation of the evidence as a whole. First, Fornataro identified two different people as the gunman in two different throw downs. According to the officer, his first identification— which he later repudiated — came in little more than half a minute, while his second identification — of defendant — came only after he had studied the pictures for two and a half minutes, suggesting that he had some difficulty with the identification. Second, we have reviewed the composite sketch of the gunman; in contrast to the dissent, we believe that whether the sketch resembles defendant is at the least [183]*183subject to reasonable debate.11 Third, Fornataro stated that the gunman was about his height. Defendant is significantly taller than Fornataro, while Sandborn is about the same height. Fourth, the gunman used a phrase that Sandborn, but not defendant, was known to use.
Most significantly, in our previous opinion we did not fully recognize the effect of Sandborn’s statements on the trier of fact’s evaluation of the evidence. If the trier of fact believed those statements, they were sufficient in themselves to lead to defendant’s acquittal of the' robbery and kidnapping charges. Sandborn, who unquestionably participated in the burglary, described what had happened almost immediately afterwards and in the process described himself as doing the things with which defendant was charged. When viewed in the light of those statements, the other evidence against defendant on the robbery and kidnapping counts loses much of its weight and there will almost certainly be a reasonable doubt about defendant’s guilt of those charges.12
As the dissent points out, it is not certain that the trier of fact would believe Sandborn’s statements. The problems that it identifies, however, are those that are inherent in those kinds of statements. A person who has just committed a crime is unlikely to talk about it to passers-by whom he or she meets on the street; rather, the person will do so to friends and close associates. The inevitable result is that evidence of the statements will have to come from those friends and associates. The differences in the witnesses’ memories about the details of Sandborn’s statements are not surprising or unusual. Indeed, they may suggest that the witnesses are remembering an actual event rather than recounting a story that they made up — and carefully coordinated — for use at the trial. Those are all issues for the trier of fact’s consideration; [184]*184they are not sufficient to affect our analysis of whether the evidence could have affected the outcome of the trial.
We turn to the Cunningham inquiries, beginning with the relative strength of the parties’ evidence. As we have noted, Fornataro’s identification of defendant was clear and unambiguous; the trial court expressly found him to be a credible witness. In addition, Fornataro had purposely studied the gunman’s face in order to be able to identify him later, but in defendant’s version of events he would have had little or no opportunity to see defendant before viewing the second throw down. On balance, the eyewitness evidence against defendant is strong. The problems that we identified in the previous paragraph are not necessarily unusual in any eyewitness identification, and they may have had little effect on the strength of the case in the absence of Sandborn’s statements.
The second Cunningham inquiry concerns the significance of the excluded evidence in light of the totality of the parties’ evidence. Here, the evidence that Sandborn, immediately after the burglary, both described the identical actions that Fornataro later described and said that he was the one who did them is potentially powerful evidence. It both confirms Fornataro’s story about the burglary and raises questions about his identification of the gunman. That is particularly the case because Sandborn’s testimony is consistent with the physical evidence concerning the location of defendant’s pager. In light of Sandborn’s statements, the problems with Fornataro’s identification take on new significance, potentially creating a reasonable doubt about its accuracy that was not otherwise present.13 A reasonable doubt about the accuracy of Fornataro’s identification would translate to a reasonable doubt about defendant’s guilt. We are mindful of the fact that the credibility of Boyd and Byles is critical to the weight to be given to Sandborn’s statements. Nevertheless, the credibility of a witness is generally within the exclusive province of the factfinder, and rightly so. In the absence of any preruling finding by the trial court that Boyd and Byles [185]*185were not credible, we are unable to conclude that their veracity is not an issue for the factfinder. Thus, in the consideration of the totality of the evidence, the excluded evidence was potentially of great significance. Its exclusion was not harmless.
We do not mean to suggest that Sandborn’s statements would have changed the result. Rather, the statements would have sufficiently changed the mix of the evidence so that they could have done so. In light of that conclusion, we certainly cannot say, as Joslin requires, that there was little, if any, likelihood that the error affected the verdict. We therefore reverse the robbery and kidnapping convictions and remand for a new trial.
Reconsideration allowed; former opinion withdrawn; convictions for robbery and kidnapping reversed and remanded for new trial; otherwise affirmed.